Skip to main content

Table 7 Relative bias in the three-component mixture model

From: Segmentation and intensity estimation for microarray images with saturated pixels

Parameter GMM0 CGMM GMM1
  10 40 70 10 40 70 10 40 70
π1 = 0.7 0.0011 0.0010 0.0011 0.0011 0.0011 0.0012 0.0006 -0.0002 -0.0088
μ1 = 2, 000 0.0015 0.0015 0.0015 0.0015 0.0015 0.0015 0.0015 0.0014 0.0033
σ1 = 1, 000 -0.0044 -0.0045 -0.0045 -0.0044 -0.0045 -0.0044 -0.0051 -0.0077 -0.0117
π2 = 0.1 -0.0022 -0.0017 -0.0023 -0.0032 -0.0032 -0.0046 0.0176 0.1637 0.5294
μ2 = 15, 000 0.0008 0.0005 0.0004 0.0003 -0.0001 0.0003 0.0117 0.1679 0.7195
σ2 = 6, 000 -0.0262 -0.0249 -0.0255 -0.0276 -0.0271 -0.0285 0.0100 0.3923 1.8881
μ 3 -0.0001 0.0000 -0.0002 -0.0002 0.0002 -0.0010 -0.0053 -0.0384 -0.1352
σ 3 -0.0071 -0.0076 -0.0071 -0.0049 -0.0028 -0.0033 -0.1161 -0.4601 -0.9294
  1. Simulation with true K = 3: Relative bias based on runs with K correctly selected by BIC. The models considered were regular Gaussian mixture for complete, uncensored data (GMM0), censored Gaussian mixture for censored data, and regular Gaussian mixture for censored data (GMM1). Percents of saturated foreground pixels were set at 10% (μ3 = 52, 700, σ3 = 10, 000), 40% (μ3 = 62, 500, σ3 = 12, 000) and 70% (μ3 = 75, 000, σ3 = 18, 000).